Slashdot Mirror
The Mainframe Is Dead! Long Live the Mainframe!
HughPickens.com writes The death of the mainframe has been predicted many times over the years but it has prevailed because it has been overhauled time and again. Now Steve Lohr reports that IBM has just released the z13, a new mainframe engineered to cope with the huge volume of data and transactions generated by people using smartphones and tablets. "This is a mainframe for the mobile digital economy," says Tom Rosamilia. "It's a computer for the bow wave of mobile transactions coming our way." IBM claims the z13 mainframe is the first system able to process 2.5 billion transactions a day and has a host of technical improvements over its predecessor, including three times the memory, faster processing and greater data-handling capability. IBM spent $1 billion to develop the z13, and that research generated 500 new patents, including some for encryption intended to improve the security of mobile computing. Much of the new technology is designed for real-time analysis in business. For example, the mainframe system can allow automated fraud prevention while a purchase is being made on a smartphone. Another example would be providing shoppers with personalized offers while they are in a store, by tracking their locations and tapping data on their preferences, mainly from their previous buying patterns at that retailer.
IBM brings out a new mainframe about every three years, and the success of this one is critical to the company's business. Mainframes alone account for only about 3 percent of IBM's sales. But when mainframe-related software, services and storage are included, the business as a whole contributes 25 percent of IBM's revenue and 35 percent of its operating profit. Ronald J. Peri, chief executive of Radixx International was an early advocate in the 1980s of moving off mainframes and onto networks of personal computers. Today Peri is shifting the back-end computing engine in the Radixx data center from a cluster of industry-standard servers to a new IBM mainframe and estimates the total cost of ownership including hardware, software and labor will be 50 percent less with a mainframe. "We kind of rediscovered the mainframe," says Peri.
IBM brings out a new mainframe about every three years, and the success of this one is critical to the company's business. Mainframes alone account for only about 3 percent of IBM's sales. But when mainframe-related software, services and storage are included, the business as a whole contributes 25 percent of IBM's revenue and 35 percent of its operating profit. Ronald J. Peri, chief executive of Radixx International was an early advocate in the 1980s of moving off mainframes and onto networks of personal computers. Today Peri is shifting the back-end computing engine in the Radixx data center from a cluster of industry-standard servers to a new IBM mainframe and estimates the total cost of ownership including hardware, software and labor will be 50 percent less with a mainframe. "We kind of rediscovered the mainframe," says Peri.
Besides the price, I'm always on the fence regarding IBM's approach to licensing. On one hand it feels like having an itemized bill with individual licenses and fees for everything down to individual screws gives more control to the buyer (as opposed to a "bundle" where one could feel like he's paying for stuff he doesn't need), but in my experience it's almost impossible to seriously weed out (or even understand) items from the list.
My best billing experience has been in a small business that was using Dell's financing. No big upfront cost, a simple monthly amount to pay. Need one more server or ten more workstations? No problem, the stuff is delivered and the monthly amount is increased by $200. Awesome.
lucm, indeed.
Have you seen those beasts? They come with earthquake kits (hydraulic suspension, gyros, etc), waterproof cables connectors (to keep working in a small flood) and nitrogen-rich fire-resistant enclosures. Drives are snapped in a backplane because loose cables are a liability, and IBM even provides an optimal distribution of redundant components inside the case based on their extensive records of hardware failures experienced by all their large customers in the last 20 years (because of course those machines are not serviced by the customers themselves).
This kind of big iron is definitely not a pimped pizza box. It is an amazing piece of engineering. Loud, expensive, inflexible, but truly amazing.
lucm, indeed.
Mainframes are like really big industrial cars where everything is hugely expensive. They're stupid expensive, but far cheaper than trying to do massive amounts of work with thousands of pickup trucks.
It's like the transporter they use to move the space shuttle with rockets and all ready to go:
http://en.wikipedia.org/wiki/C...
It goes 1MPH, which sounds pretty wuss-tastic in car terms, until you consider how much capacity it has at that speed. It would be basically impossible to accomplish the same thing with any number of VW Beetles without spending years taking apart and reassembling everything each time you wanted to attempt a launch.
That's where mainframes make sense - problems which are really massive, but need to run on one computer. Any problem that can be broken down into smaller chunks can be solved much more efficiently with a network of smaller computers.
As the smaller computers continue to get more and more capable and the technology to break down problems and high speed interconnects become more common, the jobs that run better on a mainframe get more rare and networks of servers become more common.
Mainframes do have one cool thing going for them that is not respected on smaller machines - portability. There's code that's been in use for several decades on mainframes running in a stack of emulators. Each new mainframe gets an emulator to make it possible to act just like an an old mainframe. This means the customer needs to run their code on the emulator instead of having to tweak the code to work on the new mainframe. For jobs that justify mainframe costs, downtime is very expensive, so minimizing additional conversion efforts is huge. Also, it's entirely possible that the last person who knew how some mission critical code worked may have died 40+ years ago and business people aren't big proponents of hiring someone to figure out and rewrite legacy stuff.
I'd be lying if I said it was easy and the architecture we have for the moment (temporary) is terrible as we basically did a giant recompile for the majority of the cobol code and have it running on a single server (though it uses at least 70 cores of that capacity for most of the day), long term it will be recoded with part of the savings from the mainframe decommissioning used to re-architect it to be a more scale out rather than scale up design. we had to extract all the batch, nightly analytics, archiving, forensics and various other processes on to separate machines recode some processes to be more parallel. 2 year project all up (plus a great deal of planning before that). it actually runs on 6 of those monster specced machines, only one of them is really ever stressed though, the rest are there for redundancy, testing and keeping a lot of the miscellaneous tasks off the core machine.
"We kind of rediscovered the mainframe," says Peri.
Everytime IBM announced a new mainframe line of products they hired an "external" consultant to say exactly the same bullshit. Since 1990, they announced the rediscovery of the mainframe at least five times. IBM is addicted to the mainframe given the large chunk of revenues associated to it. So, they serve us the same marketing bullshit each time.
Achille Talon
Hop!
And yet, if you open up a mainframe, you will see that on the inside, it is exactly a vastly distributed system with thousands of nodes.
No it isn't. Even this latest monster doesn't have that many actual processors in it.
The main advantage to a mainframe is its ability to shovel around vast amounts of data very rapidly. IBM has offloaded a lot of the I/O work onto the peripheral data controllers ever since the System/360.
Technologically, mainframes are lagging. This is the first IBM mainframe that has had the ability to run multiple instructions at once on a single core the way Intel chips have done for many years now. The processor clock speed isn't anything outstanding for the day, either.
It really sounds a lot like the beginning of the end. There's a lot of interesting stuff that IBM has done to their big iron systems over the years, but it's pretty much stuff that didn't transport outside their own little world. Within that world, you have all sorts of interconnects, and one should never underestimate the benefits of having One Big Box when it comes to power consumption and real-estate needs, especially since IBM's reputation has always been that just because you have One Big Box doesn't mean Single-Point failure.
But in the end, I think they may simply fade away. There's no cheap way to get into the mainframe business, The closest thing to open-source is the Hercules emulator, but the licensing fees for any IBM OS release past 1986 mean that small businesses cannot leverage it. There are all sorts of specialized skills required that are no longer dime-a-dozen. Most software products and systems that run on mainframes have counterparts that run on commodity hardware and OS's - often cheaper, and considering what IBM has done to their workforce, often better supported.
So if you have lots of legacy code to support, or are willing to dedicate a lot of expensive resources to a totally-packaged system, this new box may be wonderful for you. For the computing world at large, it's likely to be hardly noticed.
The mainframe people I know, when they rarely refer to transactions, have a slightly different meaning from when windows or unix people do it. The mainframe people more often rever to messages, which is a whole discrete task, which can often require multiple database transactions, some computational passes etc. They usually talk about hundreds of thousands of messages per hour, so if it's 2.5 billion mainframe-style "transactions"(messages), it's pretty damn impressive.
I disagree that they're inflexible. Capacity On Demand (COD) gives customers ability to pay for additional capacity (engine/CPU) only while that additional capacity is turned on. A mainframe is typically partitioned into LPARs at bare metal using PR/SM. You can add/remove/rearrange LPAR configurations. Then there is z/VM -- this is IBM's software crown jewel of mainframe software. This is where most shops run virtualized Linux servers. Create, start, stop, reconfigure guests as needed. You can reallocate storage among guests very easily. You can create software-only (virtual) networks for the guests. IBM's latest version of z/VM supports OpenStack. Many other features, these are just the main ones that come to mind. I don't call this inflexible.
The reason that this is the first mainframe with SMT is simple. Prior to the previous generation (z12), most mainframe workload was z/OS, and z/OS has no support for SMT. Starting with z12, a whole lot of mainframes started being used for new workload (Linux). Now it makes sense to add SMT, so they did. It has nothing to do with 'technologically lagging'.
As for clock speed being 'not outstanding', looking around Intel's site for server chips I don't see anything clocked above 3.4GHz. This new mainframe runs at 5GHz (previous generation was 5.5GHz, but the new one is still faster).
The 'cheap way' to get into the mainframe business is Linux, and many companies are doing it.
The reasons customers are running Linux on mainframes is for the same reasons they run anything on mainframes. In many cases, it is just a better value. The 'legacy is the only reason for mainframes' mantra is really old and tired, and is only repeated by people who know very little about mainframes.
I've seen situations where trying to replace a mainframe with a server ended in bitter failure and hundreds of thousands of dollars of expense. We're talking batch processing millions of records on the Mainframe in a few minutes, while a server managed 30,000 in a day. Sometimes, the mainframe just has better hardware.
Mainframes are designed to take hardware and software upgrades without interrupting software processes. If we ever implemented Linux's user space APIs on Minix 3 (e.g. kevent, iptables), we could run udev, dbus, and other Linux-specific subsystems on a microkernel; it would be similar to a mainframe, in that you could upgrade the core OS without rebooting, yet dissimilar, in that it wouldn't be a virtualized cluster like OpenMOSIX in which the applications move onto another running OS when you want to reboot one VM.
Security policies on the mainframe are different than PC, too. High security means each application is so isolated as to effectively run in its own VM, from a practical standpoint. From a technical standpoint, the OS is just so good at confining applications to what they're allowed to do (and those privileges are so well-defined) that it achieves similar isolation to running in separate VMs. This drastically reduces down time. Some effort has gone into Linux on the GrSecurity side to apply kernel write-execute separation; and, again, Minix 3 or a similar OS could create strict memory policies to prevent drivers from accessing kernel RAM not related to the driver and the process invoking it; this plus process groups and containers (as in Linux) and mandatory access control policies would come close, if not parity, a mainframe.
I have enough understanding to know what must be done to create something, but not how. If I knew how, I'd have long ago added services to Minix 3 to run Linux desktop subsystems for systemd, udev, and dbus; created a policy manager which can define application access policies by contexts, user, and the user's container policy (e.g. Pidgin can access the user's configured $HOME/.pidgin/ and $HOME/download/pidgin/ for read-write, etc.); and modified some of the interfaces to store data relevant only to specific processes in separate pages, and only map those pages in the appropriate context, so that a bug writing all over memory would have limited-scope damage even in kernel (this is hardly ever an issue in Minix to start with). But nay.
Support my political activism on Patreon.
I think more people will start running their own small servers. Cheap storage, always-on internet, dynamic DNS, better software. It's what I do. I have a NAS and OwnCloud and sync all my mobile stuff to that. It's all the benefits of having your data always accessible without the drawbacks of turning over your files to a 3rd party.
We don't have a state-run media we have a media-run state.